Selective Profiling of Proteins in Lung Cancer Cells fromFine-NeedleAspirates byMatrix-Assisted Laser DesorptionIonizationTime-of-Flight Mass SpectrometryJosephM. Amann,1,6 Pierre Chaurand,2 Adriana Gonzalez,3,6 James A.Mobley,2

Pierre P. Massion,4,6 David P. Carbone,5,6 and RichardM. Caprioli2,6

Abstract Purpose:Matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOFMS) has been used to analyze tumor sections and can determine tumor type, nodal involve-ment, and survival, and shows promise in predicting therapeutic response. Our purpose was todevelop a method compatible with MALDI-TOFMS that allows selective analysis of cancer cellsinmixed clinical samples such as fine-needle aspirates.Experimental Design: Lung cancer cell lines were cytocentrifuged onto metal-coated, trans-parent glass slides and used for optimization of fixation, staining, and RBC lysis protocols. Fine-needle aspirates from human tumors and mouse model tumors were used to provide fresh tissuesamples for determining the feasibility of this method.Results: The MALDI-TOF MS compatible fixation and staining techniques provided high-resolution cellular morphology, which allowed identification and selective spotting of tumor cells.The RBC lysis step efficiently removed contaminating RBC yielding spectra nearly free fromhemoglobin peaks. Protein profiles of fine-needle aspirates were found highly reproducible andsimilar to the profiles of the tissue from which they were obtained. Using this method, we wereable to differentiate between xenograft tumors derived from two different human cell lines, A549and H460.Conclusion:This procedure results in the production of high-quality, cancer cell ^ specific pro-tein profiles. This highly reproducible technique could be applied to many other types of mixedclinical samples and has the potential to be very useful in the clinical diagnosis, classification,and, potentially, the individualized treatment of cancer patients.

Whereas the treatment of lung cancer has seen significantadvances recently, the methods used for lung cancer diagnosishave not changed in decades. After identifying a suspicious massby some form of imaging, either chest X-ray or computedtomography scan, a tissue sample from the suspect lesion is

stained with H&E and examined by a pathologist under a lightmicroscope. Based on the observed cytologic features, the tumoris classified as non–small-cell lung carcinoma or small-cell lungcarcinoma. Whereas non–small-cell lung carcinoma can befurther subcategorized into three main groups, adenocarcino-ma, squamous cell carcinoma, or large cell carcinoma (1), all ofthese subtypes are treated the same in spite of a large variabilityin clinical outcomes. Clearly, practical new technologies need tobe investigated for improving diagnosis, optimizing treatmentefficacy, and reducing treatment toxicity.Matrix-assisted laser desorption ionization time-of-flight

mass spectrometry (MALDI-TOF MS) is a technology that canbe used to generate protein signatures from tissue sections thatcan contain several hundred protein signals (2). In direct tissueprofiling experiments, thin tissue sections are cut from freshfrozen biopsies, thaw-mounted on a flat, metal MALDI targetplate, and spotted with matrix (typically sinapinic acid). Totarget the cancerous tissue in the sample, an adjacent tissuesection that has been mounted on a glass slide and H&E stainedis often used as a rough guide for matrix deposition. With thismethod, protein profiles have been obtained from tumor tissuesections and successfully analyzed with a class predictionmodel developed for handling the large data sets produced byMALDI (3–5).Many clinical samples are fine-needle aspirates that result in

a liquid suspension of cells highly contaminated with blood,

Imaging, Diagnosis, Prognosis

Authors’ Affiliations: 1Department of Cancer Biology; 2Mass SpectrometryResearch Center and the Department of Biochemistry; 3Department of Pathology;4Division of Allergy, Pulmonary, and Critical CareMedicine,Veterans Affairs MedicalCenter ; 5Division of Hematology/Oncology, Department of Medicine; and6Vanderbilt-Ingram Cancer Center,Vanderbilt University, Nashville,TennesseReceived 2/7/06; revised 5/23/06; accepted 6/20/06.Grant support: NIH Lung Cancer Specialized Program of Research Excellencegrant P50CA90949 (D.P. Carbone), NIH grant GM 58008-08 (R.M. Caprioli), andNational Cancer Institute grants CA 86243-03 (R.M. Caprioli) and CA 116123-01(P. Chaurand).The costs of publication of this article were defrayed in part by the payment of pagecharges.This article must therefore be hereby marked advertisement in accordancewith18 U.S.C. Section1734 solely to indicate this fact.Note: Supplementary data for this article are available at Clinical Cancer ResearchOnline (http://clincancerres.aacrjournals.org/).J.M. Amann and P. Chaurand contributed equally to this work.Requests for reprints: David Carbone, Division of Hematology/Oncology,Department of Medicine, 685 PRB,Vanderbilt University, Nashville, TN 37232-6307. Phone: 615-936-3524; Fax : 615-936-3322; E-mail: d.carbone@vanderbilt.edu.

F2006 American Association for Cancer Research.doi:10.1158/1078-0432.CCR-06-0264

www.aacrjournals.orgClin Cancer Res 2006;12(17) September1, 2006 5142

normal cells, and noncellular debris that are smeared or spundown on slides for analysis without adjacent sections toanalyze. To accomplish light microscopic and mass spectro-metric analysis of the cells of interest in such a clinical sample,there are two essential requirements that must be met: first, it isnecessary for the sample plate to be optically transparent andconductive (for use in the MALDI apparatus); second, a stainmust be found that is simultaneously revealing of cellulararchitecture and compatible with MALDI-TOF MS.Recently, indium-tin oxide–coated transparent glass slides

have been used to mount tissues for MALDI analysis (6, 7).These slides are coated with a 160- to 240-A-thick layer ofindium-tin oxide, which provides the surface necessary formaintaining stable accelerating potential within the source ofthe time-of-flight mass spectrometer. These conductive glassslides have been found ideal to mount tissue sections with thebenefit that the sections can be viewed by light microscopybefore MS analysis. Unfortunately, the most common stainingmethods used in histology, DiffQuik and H&E, have beenfound to be incompatible with MALDI MS (8, 9). To improvevisualization of features, several staining protocols have beensuccessfully used, which include methylene blue and cresylviolet (7). These provide excellent histology with highresolution of tissue architecture and allow high-quality massspectra to be obtained.MALDI-TOF MS protein profiling combines speed, simplic-

ity, reproducibility, and sensitivity, making it a very powerfultechnology. The technological innovation introduced by theuse of indium-tin oxide–coated glass slides and MALDI-compatible stains increases this versatility. We have sought totake advantage of these features in designing a new diagnostictool for the direct analysis of low-abundance cancer cellscontained within fine-needle aspirates. We show here that bycombining a very simple existing technology, the cytocentri-fuge, which is present in every clinical setting, with thesimplicity and power of MALDI-TOF MS, a cutting edgediagnostic for lung cancer can be created that should providea wealth of information and may ultimately have a significantimpact on treatment decisions, and potentially, patient out-comes.

Materials andMethods

Cell culture. Human non–small-cell lung carcinoma cell lines(NCI-H1703, NCI-H2009, NCI-H1299, NCI-H460, and A549) andMC26 cells, a murine colon carcinoma cell line, were grown in RPMI(Invitrogen, Carlsbad, CA), 10% fetal bovine serum (Hyclone, Logan,UT), 2 mmol/L glutamine, with 10,000 units/mL penicillin and 10,000units/mL streptomycin (Invitrogen). Lewis lung carcinoma cells weregrown in DMEM, 10% fetal bovine serum, 2 mmol/L glutamine, with10,000 units/mL penicillin and 10,000 units/mL streptomycin.

Cytocentrifugation, fixation, and staining. Cells were washed twicewith 10-mL PBS, detached from the bottom of the dish by scraping, andcounted. Cells were diluted with PBS to 200,000 cells/mL and, using adouble cytofunnel (Thermo Electron, Waltham, MA), 200 and 400 AL ofthe cell suspension were cytocentrifuged (described below) ontoindium-tin oxide–coated glass slides (Delta Technologies Ltd., Still-water, MN) in a Cytospin3 (Thermo Electron) at 800 rpm for 5 minutes.Erythrocyte lysis buffer (Qiagen USA, Valencia, CA) was used to removeRBC from 60,000 lung cancer cells in 200 AL of normal saline alone,or with 2, 5, and 10 AL of whole blood. Initially, cells were fixed inacetone (Sigma, St. Louis, MO), methanol (VWR, West Chester, PA),

or 95% ethanol for 5 minutes after cytocentrifugation. It was laterdetermined that fixation in 95% ethanol provided high-quality cellmorphology and was used for all later studies. The cells were stainedwith 0.5% cresyl violet or 0.15% methylene blue for 1 minute, followedby two 1-minute washes in 70% ethanol (7). Slides were air-dried beforethe addition of matrix. Samples not analyzed immediately were storedat �80jC until the day of analysis when matrix was added. TheH&E-stained slides were prepared using standard techniques.

Simulated and clinical fine-needle aspirates. Tumors were generatedby injection of 1 � 106 MC26 murine colon carcinoma cells into theflanks of BALB/c mice or by injection of 1 � 106 NCI-H460 cells intothe left flanks and 1 � 106 A549 cells into the right flanks of nude mice.

After 30 to 35 days, resulting tumors were removed and placed into ice-cold saline. A 23-gauge needle attached to a 5-mL syringe was insertedinto the tumor. Aspirate was placed into 0.2 mL of ice-cold saline. Forhuman samples, a 25-gauge needle attached to a 5-mL syringe was

inserted into surgically resected tumor tissue or into adjacent normaltissue contained within the resection specimen immediately afterremoval of the tumor. As above, the aspirate was placed into ice-coldsaline. If the volume of saline was >0.2 mL, the sample was centrifuged

at 400 � g and the pellet resuspended in 0.2 mL of ice-cold saline. Truefine-needle aspirates were obtained from the clinic in 1 to 2 mL ofsaline on wet ice without identifiers. Samples were spun down andresuspended in 0.2 mL of ice-cold saline for further processing as

described. All samples were processed at the time they were received.Red cell lysis. Cells with whole blood or fine-needle aspirates in 200

AL of normal saline were placed on ice. Five volumes of erythrocyte lysisbuffer (Qiagen USA) were added and the samples were mixed byinversion and placed on ice for 15 minutes with occasional mixingevery 3 to 4 minutes. Cells were centrifuged at 400 � g for 5 minutes at4jC and the supernatant removed with a pipette. The samples werethen resuspended in 200 AL of normal saline for cytocentrifugation.

MALDI-TOF MS. For the cell lines, 300 nL of matrix, sinapinic acid

(3,5-dimethoxy-4-hydroxycinnamic acid; Fluka, Buchs, Switzerland) at

20 mg/mL diluted in a solution of 50% acetonitrile, 0.1% trifluoro-acetic acid (Fisher Scientific, Pittsburgh, PA), were deposited in a

randomly chosen spot within the circle of cells laid down on the slideby cytocentrifugation. This was followed by a second 300 nL of matrix

placed directly on top of the first spot to increase crystal density. For

the simulated fine-needle aspirates, clusters of cancerous cells, 100 cells(or fewer) to several hundred cells, identified by a pathologist were

hand-spotted while visualizing the cells under a microscope at �10 or�20 magnification. MALDI MS analyses were done in an Applied

Biosystems Voyager DE-STR time of flight mass spectrometer (Applied

Biosystems, Framingham, MA) equipped with a 337-nm nitrogen laseroperated at 20 Hz. Samples were analyzed in the linear mode geometry

in delayed extraction conditions under 25-kV accelerating potential.

The delayed extraction variables were optimized for maximumresolution at fm/z 15,000. Data were recorded in the m/z range of

2,000 to 70,000 by averaging signals from four independentacquisitions of 250 laser shots each (1,000 shots total). The MALDI

MS acquisition method was first calibrated using a mixture of standard

calibrating proteins: insulin (MW 5,777.6), cytochrome c (MW12,360.1), apomyoglobin (MW 16,951.6), and trypsinogen (MW

23,982). Individual spectrums were then internally calibrated usingpreviously identified mass signals. These are, for mouse samples:

cytochrome c oxidase (MW 5,443.4), histone H4 (MW 11,306), histone

H2A.2 (MW 14,004), and h hemoglobin when detected (15, 617); andfor human samples: cytochrome c oxidase (MW 5,356.2), histone H4

(MW 11,306), histone H2A.2 (MW 14,005), and a and h hemoglobinwhen detected (MW 15,126 and 15,867, respectively).

Tissue section analysis. Ten-micron-thick tissue sections from fresh-frozen tumor biopsies were cut at �15jC using a cryostat (CM3050 S,Leica Microsystems AG, Welzlar, Germany) and thaw-mounted on theconductive glass slides. Serial sections were also cut for subsequent H&Estaining. The sections were dried at room temperature before beingsuccessively rinsed in 70% and 95% ethanol for 30 seconds,

TowardMSProfiling of Clinical Samples

www.aacrjournals.org Clin Cancer Res 2006;12(17) September1, 20065143

respectively. MALDI matrix was hand-deposited on the sections bydouble spotting two 300-nL drops using an automatic pipette. MALDIMS analyses were done as described above.

Statistical analysis. Whole spectrum analysis was done as previouslydescribed (10) with the exception that classification was not carried outhere. Briefly, all mass spectra were exported as ASCII text files andimported into ProTS Data version 1.1 (Biodesix, Steamboat Springs,CO) for baseline correction, normalization by total ion current, andaligning in batch mode. The processed text files were then imported intoa script written in Matlab (Jeremy Norris, PhD, Vanderbilt University) tobe aligned according to a single m/z column. A standard weighted meanaveraging algorithm was then applied (11). In this way, m/z values werefiltered according to the highest weight that best differentiated thenormal versus cancer groups. Further filtering was carried out to excludevalues with weighted mean averages <1.0 (ref. 10; similar in respect to2 j from the mean control value) and to exclude mean intensitydifferences that fell below 2-fold (ref. 2; experimentally derived cutoffvalue often applied for tissue profiling; data not published). The filteredvalues were then used for peak detection and further evaluated byplotting the whole spectra as compared with the difference spectra inOrigin 7.0. Intravariant error and graphs depicting variance throughoutthe spectra were prepared and plotted with tools provided by Biodesix.The hierarchical clustering analysis and subsequent icicle plot werecarried out with Statistica 6.0 (StatSoft, Tulsa, OK).

Results

MALDI-TOF MS on cytospins generates a high-quality signalfrom fine-needle aspirate-sized clusters of cancer cells. Due tothe limited sample size and high degree of nontumorcontamination of most fine-needle aspirates, it is imperativethat the method be sensitive enough to detect small numbers ofcells. The sensitivity of the technique has been shown in onerecent study where protein signals were obtained from as few as10 cells isolated by laser capture microdissection (9). In ourpilot experiments to determine the feasibility of the technique,we found that it was possible to get abundant, high-quality MSsignals from 50 to 100 cells when very low volumes of matrix(f50 nL) were deposited on the cells using pulled glasscapillaries. Given the quality of the MS data recovered with sofew cells and considering that cancer cell clumps contained inneedle aspirates are frequently in this range, we decided thatpursuing the development of methods for analysis of cancercells in fine-needle aspirates was feasible.Development of cell fixation and staining techniques compatible

with MALDI-TOF MS of cytospins. For optimal morphologicidentification, fixation and staining is required. Therefore, wetested three standard fixation techniques including 95%ethanol, methanol, and acetone for compatibility with MALDIMS. Forty thousand A549 lung cancer cells were cytocentri-fuged onto an indium-tin oxide–coated slide, fixed in organicsolvent, air-dried, and profiled by MALDI MS (Fig. 1A).Ethanol and methanol showed improved MS data quality withrespect to a nonfixed control with higher ion yield (on average,by a factor of f2 in the m/z range 4,000-60,000) and slightlybetter resolution (e.g., see signal cluster at fm/z 18,000 inFig. 1A). Acetone fixation gave somewhat lower yield andresolution spectra with respect to the control. It was deter-mined that 95% ethanol fixation was the simplest procedurethat produced high-quality spectra and this was used forsubsequent experiments.Staining is necessary to visualize the cells using light

microscopy and for the identification of cell types in complex

mixtures from clinical fine-needle aspirates. Following cytocen-trifugation onto indium-tin oxide–coated glass slides, NCI-H2009 cells were fixed in ethanol and stained with cresyl violetor methylene blue. Examination of the cells by light microscopyrevealed that the nuclei could be easily seen with well-definednucleoli using either stain, but the resolution of nucleoli wassomewhat better with cresyl violet (Fig. 1B). Unlike the stainingof unfixed tissue where cresyl violet and methylene blue werecomparable (ref. 7), the spectra obtained from the fixed cellsstained with cresyl violet showed slightly better resolution andsignal-to-noise ratio than spectra from cells stained withmethylene blue (Fig. 1C). Therefore, cresyl violet was chosenfor further studies.Erythrocyte removal is a critical step for obtaining MALDI-TOF

MS spectra from clinical needle aspirates. Contamination ofclinical samples with blood and other cell types, such aspolymorphonuclear cells, macrophages, and dendritic cells, iscommon and a potential problem for MALDI MS proteinprofiling. The presence of large amounts of hemoglobin fromRBC suppresses any other protein signatures in the sample. Toreduce the hemoglobin content in fine-needle aspirate samples,an erythrocyte lysis step was added to the protocol beforecytocentrifugation. Mass spectra obtained from NCI-H1299cells treated with erythrocyte lysis buffer showed that cellsexposed to erythrocyte lysis buffer generated the identicalspectra as untreated cells (data not shown). To test theefficiency of erythrocyte lysis buffer in removing varyingamounts of contaminating blood, 6 � 104 NCI-H1299 cellsin 0.2 mL of saline were mixed with 2, 5, or 10 AL of wholeblood. The effectiveness of erythrocyte removal is seen whencomparing cytocentrifuged, erythrocyte lysis buffer– treatedsamples to untreated samples by light microscopy (Fig. 2). Inthe untreated samples, the areas between stained H1299 cellsare filled with RBC (Fig. 2A and B). Treated samples are devoidof RBC but do contain polymorphonuclear cells after lysis(Fig. 2C and D). The H1299 cells with blood were alsomonitored by MALDI MS before and after the erythrocyte lysisbuffer treatment. The efficacy of the erythrocyte lysis step isrevealed by examining the resulting spectra from cells that werecontaminated with 10 AL of whole blood (Fig. 2E). In untreatedsamples (blue spectrum), dominant a and h hemoglobin peaksat m/z 15,127 and 15,668, respectively, doubly charged speciesat m/z 7,564 and 7,835, and hemoglobin dimers at m/z 30,253(2a), 30,793 (a + h), and 31,335 (2h) are detected. After theerythrocyte lysis step (green spectrum), in samples thatcontained 10 AL of blood, hemoglobin signals were attenuatedto a level below that of abundant cellular proteins, such ashistone H4 (at m/z 11,307 and 11,349), with almost completeelimination of peaks due to the doubly charged hemoglobinspecies and the hemoglobin dimers. Several intense peaks atm/z of 3,372, 3,443, and 3,487 are present in samples that havebeen contaminated with blood and have previously beenidentified as defensins, which are proteins present at high levelsin neutrophils (12, 13). The intense defensin peaks areindicative of the high abundance of neutrophils in the samplesafter removal of large numbers of RBC. To determine the extentto which WBC affect the spectra, 10 AL of whole blood in 0.2mL of saline were treated and the MALDI MS spectra acquiredwithout attempting to avoid matrix deposition on white cells.Whereas the defensins were again very abundant, the overallintensity, and therefore the complexity, of the protein profile

Imaging, Diagnosis, Prognosis

www.aacrjournals.orgClin Cancer Res 2006;12(17) September1, 2006 5144

Fig. 1. Optimization of MALDI-compatible fixation and stainingtechniques. A, spectra produced from tests of organic fixation ofA549 cells revealed that 95% ethanol and methanol generatedhigh-quality spectra with signal-to-noise ratiof2-fold higher thanunfixed cells. Acetone fixation induced markedly lower resolutionspectra with lower signal-to-noise ratio than unfixed cells.B, photomicrographs of cresyl violet ^ stained and methyleneblue ^ stained NCI-H2009 cells that were ethanol fixed show thatcresyl violet ^ stained cells have better resolution of nuclearstructures such as nucleoli. H&E-stained cells are presented forcomparison. C, MALDI-TOFMS spectra for unstained controlNCI-H1703 cells, cresyl violet ^ stained cells, and methyleneblue ^ stained cells.

TowardMSProfiling of Clinical Samples

www.aacrjournals.org Clin Cancer Res 2006;12(17) September1, 20065145

was not significant when compared with that of the non–small-cell lung carcinoma cell line alone (data not shown).Spectra from simulated fine-needle aspirates accurately reflect

the spectra of the matched tissue. The technique outlined inFig. 3A is rapid and technically simple. With the exception ofthe erythrocyte lysis step and the change of stains from H&E tocresyl violet, this procedure is very similar to that routinely usedin clinical laboratories. To test the usefulness of the method, weused a ‘‘simulated fine-needle aspirate’’ model that mimics aclinical fine-needle aspirate, a model used in several previousstudies (14, 15). Our initial experiments used aspirates derivedfrom mouse allograft tumors and resected human tumors. Anexample of an aspirate obtained from a resected human tumorand adjacent nontumor tissue is shown in Fig. 3B. The MALDIMS spectra obtained from the aspirates are of high quality,showing >300 distinct peaks, and comparison of tumor andnontumor shows clear differences between the two (Fig. 3C).Human tumors are very heterogeneous with large areas of

stromal tissue, blood vessels, considerable numbers of inflam-matory cells, as well as areas of necrosis (Fig. 4; SupplementaryFig. S1). In addition, changes in the tumor cells themselvesoccur as molecular changes create various subclones within thetumor. Therefore, we wanted to confirm that spectra obtainedfrom simulated fine-needle aspirates of resected human tumors

resembled those of the tumor tissue. Examination of thespectrum for a simulated fine-needle aspirate and thecorresponding tissue section revealed that the simulated fine-needle aspirate and the tumor section share many of the samepeaks (f70%; Fig. 4). Almost all of the fine-needle aspiratesamples have predominant histone signals that are present inthe tissue, but at a lower intensity. Qualitatively, fresh-frozentissue samples seem to contain more peaks, presumably as aresult of additional proteins derived from normal stromaltissue, infiltrating polymorphonuclear cells, macrophages,dendritic cells, or necrotic tissue. Interestingly, the spectrumof a fine-needle aspirate from a tumor with little necrosis andvery few infiltrating neutrophils, unlike the tumor analyzed inFig. 4, gave very low defensin peaks (m/z 3,372 and 3,443) aswould be expected (Supplementary Fig. S1). Therefore, due tothe large number of signals shared between the tumor and thefine-needle aspirate and the correlation of defensin peaks withnumbers of neutrophils, we are confident that spectra obtainedfrom a fine-needle aspirate will be highly informative andaccurately reflect the make-up of the tumor from which it isderived, although spectra derived from fine-needle aspiratesand tumor sections are clearly not completely equivalent.MALDI-TOF MS from fine-needle aspirates is reproducible. To

test reproducibility between different samples obtained on the

Fig. 2. Erythrocyte lysis buffer efficiently removes contaminatingRBC and eliminates interference from hemoglobin peaks in massspectra. Cytospin of NCI-H1299 cells contaminated with 2 AL(A and C) and10 AL (B and D) of whole blood before and aftertreatment with erythrocyte lysis buffer, respectively. Beforetreatment, areas between non ^ small-cell lung carcinoma cells(violet/purple) are packed with RBC (A and B).This is especiallyevident in the sample containing10 AL of whole blood (B).Treatmentwith erythrocyte lysis buffer effectively removes RBC leavingpolymorphonuclear cells (C and D). A to D, insets, enlargedimages to emphasize the large number of RBC present and theefficiency with which the erythrocyte lysis buffer removes them.E, MALDI-TOFMS spectra from NCI-H1299 cell line contaminatedwith10 AL of whole blood before (blue spectrum) and after(green spectrum) erythrocyte lysis buffer treatment. Hemoglobinpeaks are labeled as a and h and show as doubly chargedspecies, a2+ and h2+; singly charged species, a+ and h+; dimers,2a+, ah+, and 2h+; and trimers. Location of defensins and thehistone H4 peak at m/z 11,307 is also shown.

Imaging, Diagnosis, Prognosis

www.aacrjournals.orgClin Cancer Res 2006;12(17) September1, 2006 5146

same day, two Lewis lung carcinoma tumors were isolatedfrom mice and processed as described in Materials andMethods. From each of these tumors, 10 needle passes weremade to obtain aspirates from which 15 and 21 protein profileswere acquired, respectively. To test the variability of samplesobtained on different days, two Lewis lung carcinoma tumorswere isolated from a pair of mice 3 days later. Following thesame protocol, the number of protein profiles acquired fromthese tumors was 19 and 16, respectively. The size of theclusters varied with estimates ranging from just over a hundredcells to several thousand. Figure 5A represents three randomlypulled MALDI MS spectra obtained from distinct clumps ofcancer cells isolated from the same tumor. As can be observed,these profiles were found to be highly reproducible with littlevariation. The mean spectrum from the four tumors analyzed attwo separate time points (71 spectra in all) is shown in Fig. 5B.Confidence intervals (1 SD above and below the mean) areshown in the form of dotted lines. The coefficient of varianceover the entire mass range studied (i.e., 2-30 kDa) wasconsistent across any one experiment, ranging from 0.46 to0.69 (25th-75th inner quartiles; Supplementary Fig. S2). Theanalysis of single-day versus day-to-day experiments increasedthe geometric mean coefficient of variance value from 0.53 tof0.57. The above-mentioned results indicate that we canreproducibly harvest and prepare fine-needle aspirates forMALDI-TOF MS.MALDI-TOF MS of simulated fine-needle aspirates can be used

to discern tumor types. It was then tested whether thistechnique can generate spectra of sufficient quality to distin-guish between different tumor types. To this end, in twoseparate experiments, two xenograft tumors, derived from A549(adenocarcinoma) and H460 (large cell carcinoma) cell lines,were simultaneously grown in opposite flanks of the samemouse. After tumor removal, 10 needle passes were made to

obtain aspirates from each tumor. The aspirates were processedand matrix added as described above. In the first experiment,28 and 20 protein profiles were obtained from visuallyidentified cancer cell clumps from the A549 and H460 derivedtumors, respectively. In the second experiment, a smaller set of14 and 9 spectra were obtained for the A549 and H460 derivedtumors, respectively.From the first experiment, the resulting average spectra

shown for each tumor type clearly illustrate distinct differencesbetween the protein profiles (Fig. 6A). Within the same figure, adifference plot, shown in blue, highlights those peaks that havebeen found to be statistically different between the two tumortypes. The inset in Fig. 6A illustrates an isolated peak that wasfound to be one of the major discriminators between the twotumor types. This peak is identified in the difference plot by theblack arrow. The average peak height is shown as a solid linewith 1 SD above and below the average spectra depicted asdotted lines. It should be noted that there is no overlapbetween the SE plots, indicating a high level of significancebetween these two tumors with just a single marker.The icicle plot (Fig. 6B) depicts a qualitative supervised

Hierarchical Cluster Analysis carried out on the top 18 markersand shows a clear separation between the H460 and A549aspirates. It should be noted that two spectra taken from thesame H460 tumor were shown to fall into the A549 group.These two samples came from the same pass of the needle.Considering the high level of separation observed between thetwo tumors analyzed, this result was unexpected. Therefore, it ispossible that spectra from these two samples reflect a peculiarityin the region from where this pass was done or a spotting error,which will be discussed further. Results of the supervisedHierarchical Cluster Analysis of the second experiment gavevery similar results with 100% of the smaller set of spectra(14 from A549 and 9 from H460) being classified correctly.

Fig. 3. Procedure for processing fine-needle aspirates results in high-resolutionstaining of cell clusters and high-qualityspectra.A, the procedure is very much like astandard cytospin sample preparation withthe exception of an erythrocyte lysis step.See text for details. B, photomicrographs ofcresyl violet ^ stained cancer cells present inthe simulated fine-needle aspirate from asurgically resected human lung tumor (left)and nontumor control (right) taken fromoutside the tumor boundary. Matrix wasadded to each of these clusters of cellswhile visualizing the samples under a lightmicroscope. C, MALDI-TOFMS spectraobtained from tumor andnontumor samplesshow similarities with many peaks incommon but also distinct differences(arrows). Inset, an enlarged area fromm/z9,500 to10,700 where many peakdifferences between tumor and nontumorare apparent. Inset, peaks with the samem/z align properly (dotted line).

TowardMSProfiling of Clinical Samples

www.aacrjournals.org Clin Cancer Res 2006;12(17) September1, 20065147

Quality spectra from clinical samples. True fine-needleaspirates were obtained from the clinic without identifiersand prepared as described (Fig. 3A). Cresyl violet–stainedslides were examined by a pathologist for the presence of tumorcells. Two of the samples were positive for tumor cells andanalyzed further by MALDI-MS. The spectra obtained were richwith peaks and of high quality (Fig. 7).

Discussion

We have developed a technique based on MALDI-TOF MS forthe analysis of cancer cells contained within fine-needleaspirates, which is suitable for the clinical setting. Specimenscan be processed easily and rapidly using common reagents anda cytocentrifuge, which can be found in any cytologylaboratory. By removing contaminating blood with a simpleprocedure, tumor cells can be identified by light microscopyand selectively analyzed by MALDI-TOF MS without ionsuppression from contaminating hemoglobin.Reproducibility is prerequisite for any technology if it is to be

used for clinical decision-making. From this perspective, wefound that multiple samples derived from the same tumor,different tumors, or tumors analyzed on different daysgenerated nearly identical spectra after processing with our

novel approach. In two separate experiments, samples gener-ated from tumors that were derived from two different celllines, H460 and A549, neatly segregated into two distinctgroups following a supervised Hierarchical Clustering Analysis.This qualitative analysis clearly shows reproducibility withinany single group and further illustrates how this level ofreproducibility allows the clear separation between differinggroups of samples such as in our case two distinct tumor types.However, in one of the experiments, 96% (46 of 48) of thesamples were classified correctly. Of three cell clusters spottedfrom the same needle pass into the H460 tumor, two weremisclassified. One likely explanation for the misgrouping isthat this particular pass of the needle came from an area thatcontained cells undergoing necrosis, which gave some signals,unlike H460 or A549. Difficulties arising from sampling errorsare unfortunately intrinsic to blind needle aspirates, butemphasize the importance of attempting to obtain samplesfrom the periphery of the tumor and avoiding potentiallynecrotic areas. Alternatively, signal contamination from othercell types could possibly come from our current method ofmanually spotting clumps using a pulled capillary. Althoughsmall spots can be achieved, the placement accuracy may insome cases be relatively imprecise. Automated matrix dispens-ing robots are currently being developed with placement

Fig. 4. Fine-needle aspirates and matched tissue sections derived from the sametumor have many peaks in common. Comparison between the spectrum from thetumor tissue section (see H&E-stained section in photomicrograph) and thespectrum from the fine-needle aspirate acquired from that tumor. Bottom,m/z rangefrom 3,000 to19,000.Top, a region from4,700 to10,500 that has been expanded tomore clearly show the similarities between the tumor section and the fine-needleaspirate. Only the major peaks held in common across the two samples have beenlabeled. Inset, photomicrograph of a human tumor (adenocarcinoma), depictingthe heterogeneity that is seen in human tumor specimens. In this tumor, areas ofnecrosis (N) are surrounded by tumor cells (T). Between the tumor cells are largeareas of stroma (S). Note the large defensin peaks atm/z 3,371and 3,443 indicativeof neutrophil infiltration, most likely due to the high level of necrosis.

Fig. 5. MALDI-TOF analysis of the tumors is highly reproducible.A, comparison ofthree spectra from three different needle passes through a Lewis lung carcinomatumor shows that the spectra are essentially identical. B, the mean spectrum fromthe four tumors analyzed at two separate time points (35 spectra in all) is shownalong with the confidence intervals (dotted lines ; 1SD above and below the mean).Inset, an expanded region of the spectrum from 9,000 to10,500 with the purposeof more clearly showing the confidence intervals.

Imaging, Diagnosis, Prognosis

www.aacrjournals.orgClin Cancer Res 2006;12(17) September1, 2006 5148

accuracies in the order of 20 Am or better (16), which we intendto use as we refine the technology. Regardless, the overallresults were quite encouraging as all other spectra clusteredtightly within their respective groups. More work is currentlybeing carried out to further refine this unique and verypromising approach and to translate it into the analysis of realclinical specimens.Many technologies are currently being investigated for

refining our ability to diagnose cancer, define cancer type,and predict prognosis and response to treatment. The mostdeveloped of these is gene expression profiling using micro-array analysis. Indeed, microarray studies have been done usingfine-needle aspirates from several types of solid tumors withoutpurification of tumor from nontumor cell types (14, 15,17–20). However, problems have been encountered inisolating RNA of sufficient quality for microarray analysis andmost clinical samples will be too heterogeneous to providemeaningful results.Additional technologies include multiplex analysis of

known biomarkers by bead-based multiplexing, where anti-bodies are coupled to bead matrices, and high-density single

nucleotide polymorphism arrays (21, 22). As previouslymentioned, the heterogeneity of most clinical samplescomplicates the analysis of either technology and could affectthe result. By using the combination of erythrocyte lysis bufferto remove RBC and the indium-tin oxide–coated glass slideswith MALDI-compatible stains, we have circumvented theproblem of potentially interfering contaminants. Directlyspotting clusters of cancer cells with matrix by visualizingthem under a microscope, with few exceptions, leaves the largemajority of contaminating cells untouched and the analysisfree from such complications. The equipment to enable image-guided, robotic spotting and spectrum acquisition is currentlybeing developed and should further enhance our ability toprecisely target tumor cells.A major obstacle to progress in this field is the difficulty in

obtaining suitable numbers and types of tumor samples frompatients necessary for training the specialized algorithmsdesigned to find differential peaks. As current collectionprotocols in clinical trials change, with requirements thattumor biopsies be taken before and after treatment, thenumbers of samples available for analysis should increasedramatically, particularly if fine-needle aspirates are usable forboth defining and testing these patterns. With a large numberof diagnoses at most institutions being accomplished by fine-needle aspirates, the ability to use these specimens with thetechnology we describe here should speed our progress.Standardization of the technique will allow collaborationsbetween medical centers and allow a more rapid accumulationof the necessary data.The MALDI-TOF MS technology has the potential to fill a

significant need in the area of diagnostic medicine. Thesensitivity makes it especially suited for small sample sizes.Data can be generated with minimal training due to the ease ofthe technique. The speed of the sample preparation and dataacquisition makes the processing of large numbers of samples

Fig. 6. MALDI-TOFMS of fine-needle aspirates can be used to discriminatebetween tumor types. A, average spectra are shown comparing A549 and H460along with an associated difference plot in blue depicting the discriminatebiomarkers used to discern differences between the two tumor types (the statisticalapproach used to discern markers of interest is discussed in Materials andMethods). Inset, an isolated peak that was found to be one of the majordiscriminators used to classify the two tumor types.The average peak height isshown with1SD above and below the mean for each spectrum (dotted lines).This peak is identified in the difference plot by the black arrow. B, the icicle plotshown depicts a qualitative supervised Hierarchical ClusterAnalysis using the top18 markers and shows a clear separation between the H460 and A549 aspirates.

Fig. 7. MALDI-TOFMS analysis of clinical fine-needle aspirates is of high quality.Clinical fine-needle aspirates were obtained and processed as described in the text.The spectra are still rich in peaks throughout the presented mass despite thepresence of the hemoglobin still significantly observed in these samples even afterthe erythrocyte lysis step.

TowardMSProfiling of Clinical Samples

www.aacrjournals.org Clin Cancer Res 2006;12(17) September1, 20065149

possible. Finally, the cost of the method, including materials,technician time, and mass spectrometer time, which can beprovided by specialized reference institutions for a nominal fee,would be a fraction of the cost of other methods beingdeveloped for cancer diagnosis and response prediction. All ofthese attributes make the use of the cytospin-MALDI-TOF MStechnology for the processing of fine-needle aspirates wellsuited for potential translation into the clinical laboratory.

Acknowledgments

We thank Candace Murphy for the processing of human tumor fine-needle aspi-rate samples and Scott Hiebert for the use of their cytocentrifuge; Paul Bunn for theNCI-H1703 and NCI-H460 cell lines, JohnMinna for the NCI-1299 and NCI-H2009cell lines, and BruceJohnson for theA549 cell line; LiYang for producing theMC26tumors, Sergey Novitskiy for providing us with Lewis Lung Carcinoma tumors, andThaoDang for providing thenudemice and for growing theA549 andH460 tumorsused in this study.

References1. TravisWD, Brambilla E, Muller-Hermelink HK, HarrisCC, editor.WHO classification of tumors: pathologyand genetics of tumours of the lung, pleura, thymus,and heart. Lyon (France): IARCPress; 2004.

2. Chaurand P, Sanders ME, Jensen RA, Caprioli RM.Proteomics in diagnostic pathology: profiling andimaging proteins directly in tissue sections. Am JPathol 2004;165:1057^68.

3.Yanagisawa K, ShyrY, Xu BJ, et al. Proteomic patternsof tumour subsets in non-small-cell lung cancer.Lancet 2003;362:433^9.

4. Jamshedur Rahman SM, ShyrY,Yildiz PB, et al. Pro-teomic patterns of preinvasive bronchial lesions. AmJRespir Crit Care Med 2005;172:1556^62.

5. Schwartz SA,Weil RJ, Thompson RC, et al. Proteo-mic-based prognosis of brain tumor patients usingdirect-tissue matrix-assisted laser desorption ioniza-tion mass spectrometry. Cancer Res 2005;65:7674^81.

6. Galicia MC, Vertes A, Callahan JH. Atmosphericpressure matrix-assisted laser desorption/ionizationin transmission geometry. Anal Chem 2002;74:1891^5.

7. Chaurand P, Schwartz SA, Billheimer D, Xu BJ,Crecelius A, Caprioli RM. Integrating histology andimaging mass spectrometry. Anal Chem 2004;76:1145^55.

8. Todd PJ, McMahon JM, Short RT, McCandlish CA.

Organic SIMS of biologic tissue. Anal Chem1997;69:529^35A.

9. Xu BJ, Caprioli RM, Sanders ME, Jensen RA. Directanalysis of laser capture microdissected cells byMALDImass spectrometry. JAmSocMassSpectrom2002;13:1292^7.

10.MobleyJA, LamYW, Lau KM, et al. Monitoring theserological proteome:the latest modality in prostatecancer detection. JUrol 2004;172:331^7.

11.GolubTR, SlonimDK,Tamayo P, et al. Molecular clas-sification of cancer:class discovery and class predic-tion by gene expression monitoring. Science 1999;286:531^7.

12. Schneider JJ, Unholzer A, Schaller M, Schafer-Korting M, Korting HC. Human defensins. J MolMed 2005;83:587^595.

13. Zhang L,YuW, HeT, et al. Contribution of human a-defensin 1, 2, and 3 to the anti-HIV-1activity of CD8antiviral factor. Science 2002;298:995^1000.

14. Ellis M, Davis N, Coop A, et al. Development andvalidation of a method for using breast core needlebiopsies for gene expression microarray analyses. ClinCancer Res 2002;8:1155^66.

15. Assersohn L, Gangi L, ZhaoY, et al.The feasibility ofusing fine needle aspiration from primary breast can-cers for cDNA microarray analyses. Clin Cancer Res2002;8:794^801.

16. Aerni HR, Cornett DS, Caprioli RM. Automated

acousticmatrix deposition forMALDI sampleprepara-tion. Anal Chem 2006;78:827^34.

17. Pusztai L, Ayers M, Stec J, et al. Gene expressionprofiles obtained from fine-needle aspirations ofbreast cancer reliably identify routine prognosticmarkers and reveal large-scale molecular differencesbetween estrogen-negative and estrogen-positivetumors. Clin Cancer Res 2003;9:2406^15.

18. Sotiriou C, Powles TJ, Dowsett M, et al. Geneexpression profiles derived from fine needle aspirationcorrelate with response to systemic chemotherapy inbreast cancer. Breast Cancer Res 2002;4:R3.

19.Wang E, Miller LD, Ohnmacht GA, et al. Prospectivemolecular profiling of melanoma metastases suggestsclassifiers of immune responsiveness. Cancer Res2002;62:3581^6.

20. Lim EH, Aggarwal A, AgasthianT, et al. Feasibility ofusing low-volume tissue samples for gene expressionprofiling of advanced non-small cell lung cancers. ClinCancer Res 2003;9:5980^7.

21. Keyes KA, Mann L, Cox K, et al. Circulating angio-genic growth factor levels in mice bearing humantumors using Luminex Multiplex technology. CancerChemother Pharmacol 2003;51:321^7.

22. Zhao X, Li C, Paez JG, et al. An integrated viewof copy number and allelic alterations in the cancergenome using single nucleotide polymorphism arrays.Cancer Res 2004;64:3060^71.

Imaging, Diagnosis, Prognosis

www.aacrjournals.orgClin Cancer Res 2006;12(17) September1, 2006 5150